[PATCH v2] net/mlx5: use just sufficient barrier for Arm platforms

[Honnappa Nagarahalli]

<snip>

> 
> >
> > Hi, Honnappa
> Hi Slava, thanks for the feedback.
> 
> >
> > We discussed the barrier here:
> > http://patches.dpdk.org/project/dpdk/patch/20210606164948.35997-1-
> > honnappa.nagarahalli at arm.com/
> Yes, I have changed the patch according to the discussion. i.e. barrier is
> needed, but different (inner sharable domain) barrier is required.
> 
> >
> > (BTW, it is good practice to keep the reference to previous patch
> > versions below Commit Message of the next ones).
> >
> > This barrier is not about compiler ordering, it is about external HW
> > agent memory action completions.
> > So, I'm not sure the rte_atomic_thread_fence() is safe for x86 - patch
> > impacts
> > x86 as well.
> The earlier barrier 'rte_io_rmb()', resolves to a compiler barrier on x86 [1].
> The rte_atomic_thread_fence(__ATOMIC_ACQUIRE) on x86 also acts as a
> compiler barrier. So, there is no change for x86.
> 
> 
> [1]
> https://github.com/DPDK/dpdk/blob/main/lib/eal/x86/include/rte_atomic.h#
> L80
Hi Slava, any more comments on this?

> 
> >
> > With best regards,
> > Slava
> >
> > > -----Original Message-----
> > > From: Honnappa Nagarahalli <honnappa.nagarahalli at arm.com>
> > > Sent: Tuesday, August 30, 2022 23:01
> > > To: dev at dpdk.org; honnappa.nagarahalli at arm.com;
> > ruifeng.wang at arm.com;
> > > Matan Azrad <matan at nvidia.com>; Shahaf Shuler
> <shahafs at nvidia.com>;
> > > Slava Ovsiienko <viacheslavo at nvidia.com>
> > > Cc: nd at arm.com; Matan Azrad <matan at nvidia.com>; stable at dpdk.org
> > > Subject: [PATCH v2] net/mlx5: use just sufficient barrier for Arm
> > > platforms
> > >
> > > cqe->op_own indicates if the CQE is owned by the NIC. The rest of
> > > the fields in CQE should be read only after op_own is read. On Arm
> > > platforms using "dmb ishld" is sufficient to enforce this.
> > >
> > > Fixes: 88c0733535d6 ("net/mlx5: extend Rx completion with error
> > > handling")
> > > Cc: matan at mellanox.com
> > > Cc: stable at dpdk.org
> > >
> > > Signed-off-by: Honnappa Nagarahalli <honnappa.nagarahalli at arm.com>
> > > Reviewed-by: Ruifeng Wang <ruifeng.wang at arm.com>
> > > ---
> > >  drivers/common/mlx5/mlx5_common.h | 6 +++++-
> > >  1 file changed, 5 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/drivers/common/mlx5/mlx5_common.h
> > > b/drivers/common/mlx5/mlx5_common.h
> > > index 5028a05b49..ac2e85b15f 100644
> > > --- a/drivers/common/mlx5/mlx5_common.h
> > > +++ b/drivers/common/mlx5/mlx5_common.h
> > > @@ -195,7 +195,11 @@ check_cqe(volatile struct mlx5_cqe *cqe, const
> > > uint16_t cqes_n,
> > >
> > >  	if (unlikely((op_owner != (!!(idx))) || (op_code ==
> > > MLX5_CQE_INVALID)))
> > >  		return MLX5_CQE_STATUS_HW_OWN;
> > > -	rte_io_rmb();
> > > +	/* Prevent speculative reading of other fields in CQE until
> > > +	 * CQE is valid.
> > > +	 */
> > > +	rte_atomic_thread_fence(__ATOMIC_ACQUIRE);
> > > +
> > >  	if (unlikely(op_code == MLX5_CQE_RESP_ERR ||
> > >  		     op_code == MLX5_CQE_REQ_ERR))
> > >  		return MLX5_CQE_STATUS_ERR;
> > > --
> > > 2.17.1